Presentation (CTI)

OBJECTIVES: With sufficient data, statistical evidence of instransitivity in network meta-analysis may be detected through tests of inconsistency/incoherence, but these tests are generally underpowered. In these cases, decisions regarding potential intransitivity rely exclusively on evidence synthesis feasibility assessments that typically compare included trials on inclusion/exclusion, outcome definitions, estimands, baseline risk, and patient characteristics. The current research introduces a novel additional method to leverage expert opinion to flag potential intransitivity based on absolute outcomes estimated by the model and those at the trial level.
METHODS: A case study based on trials in moderate-to-severe plaque psoriasis is used to explore two implementations: (1) The estimated treatment effects are used to transport each treatment into every trial in the network. (2) The treatment effects in each trial on PASI 75 and PASI 90 are applied to the placebo arm of all trials to assess potential violations of ordering constraints.
RESULTS: Transporting model estimated effects into each trial allows for detection of implausibly high or low predicted absolute effects that may suggest the presence of effect modifiers. Improvement in the face validity of estimates after adjustment for differences in baseline risk can complement existing meta-regression diagnostics. Further, we find that observed treatment effects on PASI 75 and 90 would be impossible to observe in several trials based on violating the ordering constraint that the proportion of PASI 75 responders will always be equal to or greater than the proportion of PASI 90 responders. This may suggest intransitivity or the need for a model that enforces ordering.
CONCLUSIONS: Leveraging clinical expertise and/or known outcome constraints in combination with estimated absolute outcomes is a powerful tool in the detection of potential intransitivity in networks where formal inconsistency analyses are underpowered or not possible to conduct.